Reliable Wireless Indoor Localization via Cross-Validated Prediction-Powered Calibration
This work solves the problem of calibration data scarcity for researchers and practitioners in wireless indoor localization, but it is incremental as it builds on existing methods for synthetic labels and calibration.
The paper tackles the problem of reliable wireless indoor localization by addressing calibration data scarcity, proposing an approach that efficiently uses limited data to fine-tune a predictor and estimate synthetic label bias, resulting in prediction sets with rigorous coverage guarantees validated on a fingerprinting dataset.
Wireless indoor localization using predictive models with received signal strength information (RSSI) requires proper calibration for reliable position estimates. One remedy is to employ synthetic labels produced by a (generally different) predictive model. But fine-tuning an additional predictor, as well as estimating residual bias of the synthetic labels, demands additional data, aggravating calibration data scarcity in wireless environments. This letter proposes an approach that efficiently uses limited calibration data to simultaneously fine-tune a predictor and estimate the bias of synthetic labels, yielding prediction sets with rigorous coverage guarantees. Experiments on a fingerprinting dataset validate the effectiveness of the proposed method.